Status · Anthropic

Is Claude Code down?

Checking Claude Code's official status… Live · auto-refreshes every 60s
Last 30 days — live from Anthropic's incident history
incidents reported in the last 30 days
median time to resolution
last incident
Incident ledger — days with reported incidents (not an uptime %)
Jul 12 Today
Recent incidents

How to check Claude Code's status properly

"Claude Code" is its own component on Anthropic's status page — it can be degraded while claude.ai chat is fine, and vice versa.

"Claude API" covers api.anthropic.com. If it's down, Claude Code is down with it, along with every tool that calls Claude.

Model-level incidents ("Elevated errors on Claude …") often affect one model family only — switching models can be enough.

status.claude.com →

Is it down, or is it just you?

A 401 or 403 error is almost always your own auth — an expired token or a session that needs re-login, not an outage. A run of 5xx errors, or a 529 "overloaded" response, during an incident that's actually listed on the vendor's status page is on them, not you.

Before you conclude it's a real outage: try a second surface (the web app instead of the CLI, or vice versa), rule out your own network or VPN, and only then treat the vendor's official status page as the final word.

What to do while Claude Code is down

Frequently asked questions

How do I check if Claude Code is down?

Look at the live board above — it reads Anthropic's official status API (status.claude.com) directly in your browser, including the dedicated "Claude Code" component. If the board is unreachable, open status.claude.com itself.

Why does Claude Code fail while claude.ai works?

Anthropic tracks them as separate components. Claude Code depends on the Claude API and its own backend, so a component-level incident can hit the CLI while chat on claude.ai stays green.

What do 529 errors in Claude Code mean?

A 529 is Anthropic's "overloaded" response. During a listed incident it simply means capacity is constrained — retrying later or switching to a less-loaded model usually gets you moving before the incident fully resolves.